Abstract
The University of Central Florida invention describes deep-learning-based generative model methods—Generative Adversarial Networks (GAN)—for civil structural health monitoring and damage prediction. Damage diagnostic data is limited, as data collection can be costly and challenging or because there are not enough data from damaged areas to train detection models. Using information collected from multiple structures, synthetic data samples can be generated by state-of-the-art GAN models and used to train damage diagnostic systems. These tools could aid engineers, stakeholders, and decision-makers (1) to perform diagnostics accurately and effectively on raw acceleration signals and (2) to be proactive rather than reactive in managing the life cycle of structures.
Partnering Opportunity
The research team is seeking partners for licensing and/or research collaboration.
Stage of Development
Prototype available.
Benefit
Minimizes the need for dynamic response data collected from structures by generating data samples to fulfill the need of any classIncreases the performance of the AI models used for vibration-based damage diagnosticsExperiments yield 97 percent classification accuracy for synthetically enhanced datasetsMarket Application
Civil structural condition assessment servicesPublications
Generative
Adversarial Networks for Data Generation in Structural Health Monitoring, Frontiers
in Built Environment, 11 February 2022, https://doi.org/10.3389/fbuil.2022.816644.
Brochure
